Search⌘K

Pixelift/API Documentation

For AI agents

MCP server

Connect Pixelift to Claude Code, Cursor, Codex or any MCP client. The agent calls Pixelift tools directly, pays from your credit balance at the same prices as on the website, and the results land in your account.

Server address

https://pixelift.pl/api/mcp

Transport

Streamable HTTP, stateless, JSON responses

Authorization

API key in the header Authorization: Bearer <key>

Create an API key

Claude Code

Run this in your terminal with your own key:

claude mcp add --transport http pixelift https://pixelift.pl/api/mcp --header "Authorization: Bearer pk_YOUR_API_KEY"

Cursor

Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json in a project:

{
  "mcpServers": {
    "pixelift": {
      "url": "https://pixelift.pl/api/mcp",
      "headers": {
        "Authorization": "Bearer pk_YOUR_API_KEY"
      }
    }
  }
}

Other clients

Any client that supports remote MCP servers over HTTP works: use the address above and send the API key in the Authorization header.

Tools (33)

Names and descriptions exactly as the agent receives them, in English.

Tool	Cost	What it does
`pixelift_generate_image`	credits	Creates images from a text prompt with a chosen model (call pixelift_list_image_models to compare). Default nano-banana-2.1 (good all-rounder, readable text). Aspect ratios include 4:5 for Instagram/Facebook feed ads, 9:16 for Stories/Reels, 1:1 square. Write prompts like a brief: subject, setting, lighting, camera, text to render in quotes.
`pixelift_edit_image`	credits	Changes an existing image according to an instruction while keeping the subject, identity and composition (Nano Banana Pro, 2K). Use for: new background or scene for a product, removing objects, changing colors or materials, adding props, style changes. Costs 8 credits.
`pixelift_expand_image`	credits	Extends the canvas with AI-generated surroundings (outpainting). Use to turn a square product shot into a wider or taller ad format without cropping the product. For an exact ad format (1:1, 4:5, 9:16, 16:9, 1.91:1) with the original kept pixel-exact use pixelift_resize_to_format instead. Costs 2 credits.
`pixelift_resize_to_format`	credits	Converts a static ad or product image to an exact aspect ratio (1:1, 4:5, 9:16, 16:9, 1.91:1) without cropping: AI extends the background around it and the original (text, logo, product) is pasted back pixel-exact, so nothing is cut off or redrawn. Use for Meta feed (1:1, 4:5), Stories/Reels/TikTok (9:16), landscape video and YouTube (16:9) and link/Google Display ads (1.91:1). If the image already has that ratio the call returns already_in_ratio and charges nothing: use the image as is. Costs 2 credits per format. Synchronous: returns a PNG URL (max 2048 px on the long side).
`pixelift_colorize_photo`	credits	Adds natural color to a black-and-white or faded photo while keeping faces, pose and details from the original. Costs 1.5 credits. May return status running; then call pixelift_wait.
`pixelift_restore_old_photo`	credits	Removes scratches, noise and fading from an old photo and sharpens faces. Costs 3 credits.
`pixelift_upscale_image`	credits	Enlarges an image 2x, 4x or 8x with AI detail recovery. Costs 1 credit per started 4 megapixels of output.
`pixelift_remove_background`	credits	Cuts out the subject and returns a PNG with a transparent background. Costs 1.5 credits.
`pixelift_vectorize`	credits	Traces a logo, icon or illustration into a clean, scalable SVG (Recraft). Best input: a flat graphic with clear edges, up to 5 MB. Costs 2 credits. Synchronous: returns svg_url (a private file of this account, readable with the same API key for 24 hours) and the SVG size.
`pixelift_face_swap`	credits	Puts the face from face_url onto the person in image_url, keeping the pose, lighting and background of image_url (Nano Banana 2.1, Seedream as backup). Biometric data: works only for accounts that enabled biometric consent for face swap once at https://pixelift.pl/account/privacy (otherwise 403, nothing is charged), and only with photos of synthetic people or people who gave consent. Costs 1 credits. Synchronous: returns the result URL. Pairs rejected by the provider safety filter return content_flagged and are refunded.
`pixelift_colorize_video`	credits	Colorizes an old black-and-white film and keeps its sound. Costs 10 credits per clip. Asynchronous: returns a job id; call pixelift_wait until it finishes (about 1-3 minutes per 30 seconds of video).
`pixelift_generate_video`	credits	Creates a short video clip (5-10 s) from a text prompt, optionally animating image_url as the first frame (product shots, UGC b-roll, Reels). Call pixelift_list_video_models to compare models, lengths, audio and credits. Default h3-max-turbo (cheapest model with sound). If the model does not support the requested duration or aspect ratio, the nearest supported value is used. Premium models need the PLUS plan. Asynchronous: returns a job id; call pixelift_wait until it finishes (usually 1-5 minutes). Result: an MP4 URL on pixelift.pl.
`pixelift_talking_avatar`	credits	Animates a portrait photo into a talking-head video that speaks the given text (voice-over generated automatically) or an audio_url. Use for UGC-style presenters and explainers. Use only synthetic portraits (people who do not exist); the account must have biometric consent "talking_avatar" enabled once at https://pixelift.pl/account/privacy (otherwise 403 biometric_consent, nothing is charged). Costs 15 credits per started 10 seconds of speech (standard) or 25 (hd); max 30 seconds. Asynchronous: returns a job id; call pixelift_wait until it finishes (1-5 minutes).
`pixelift_text_to_speech`	credits	Generates a natural voice-over (ElevenLabs v3, multilingual) from text, up to 3000 characters. Use for ad scripts and UGC narration; pass the result URL as audio_url to pixelift_talking_avatar. Costs 1 credits per started 500 characters. Synchronous: returns the MP3 URL.
`pixelift_add_captions`	credits	Transcribes the speech in a video (Whisper) and burns styled, word-timed captions into it, ready for Reels/TikTok. Styles: clean, bold, karaoke. Costs 2 credits per started minute (minimum 3); max 5 minutes. Asynchronous: returns a job id; call pixelift_wait.
`pixelift_reframe_video`	credits	Converts a video to another aspect ratio for Reels/TikTok (9:16), Meta feed (4:5), square (1:1) or YouTube (16:9). Two modes. mode "smart" (default, recommended): crops the frame and moves the crop smoothly to follow the main subject (product, person), cutting instantly on scene cuts; audio kept. Free (0 credits), aspect ratios 9:16, 4:5, 1:1, 16:9, clips up to 60 seconds; it only crops, so nothing new is invented, and parts of a very wide shot fall outside the frame. mode "ai": Luma Reframe extends the scene with AI instead of cropping; costs 3 credits per second, max 12 seconds; use it only when the user asks for AI-extended framing. Asynchronous: returns a job id; call pixelift_wait (smart: usually under a minute).
`pixelift_edit_video`	free	Cuts and joins video clips on the Pixelift server: one clip with start/end trims it, several clips are joined in order into one MP4. Every clip is scaled (with black bars if its shape differs), resampled and re-encoded to the first clip's resolution and frame rate; clips without sound get silence, so audio stays in sync. Use it to put a new hook in front of an ad: clips = [{url: hook clip, end: 3}, {url: original ad, start: 3}]. Limits: up to 6 clips, 35 MB downloaded in total, 120 s of output. Free to call (no credits). Synchronous: returns the MP4 URL (usually 5-60 s).
`pixelift_extract_frame`	free	Saves one frame of a video as a JPG, e.g. the opening frame of an ad or the frame where the original continues after a new hook, to animate it with pixelift_generate_video (image_url) or edit it with pixelift_edit_image. Free to call (no credits). Synchronous: returns the JPG URL.
`pixelift_lipsync_video`	credits	Replaces the speech in a video of a presenter so the lips match a new audio track (Sync Lipsync 2.0 Pro). Use to localize an ad: pass the original video and either audio_url (a ready voice-over, e.g. from pixelift_text_to_speech) or text (a voice-over is generated first with pixelift_text_to_speech and charged separately). The output is as long as the shorter of video and audio, max 30 seconds. Costs 10 credits per started 10 seconds of output. Use only videos of synthetic presenters or people who gave consent; the account must have biometric consent "lipsync_video" enabled once at https://pixelift.pl/account/privacy. Asynchronous: returns a job id; call pixelift_wait until it finishes (1-3 minutes).
`pixelift_transcribe`	credits	Transcribes the speech in a video or audio file (Whisper) and returns the text with timestamped segments. Use it as the first step of ad localization: transcribe, translate the text yourself, then pixelift_text_to_speech and pixelift_lipsync_video. Costs 0.5 credits per file (up to 5 minutes). Synchronous: returns the transcript.
`pixelift_product_photoshoot`	credits	Places ONE product photo into up to 6 ready scenes for e-commerce and ads while keeping the product exactly the same (shape, label, logo, colors). Scenes: studio, marble, wood, kitchen, outdoor, dark, festive, beach, or custom (describe it in custom_scene). Mode realistic = new lighting and shadows with an automatic product-fidelity check; exact = pixel-exact product, only the background is replaced. Best input: a clean packshot of the product. Costs 1 credit per scene; failed scenes are refunded. Asynchronous: returns a session id; call pixelift_get_product_photoshoot until done.
`pixelift_get_product_photoshoot`	free	Waits up to 25 seconds and returns each scene with its status, result URL and product-fidelity verdict. Call again until done is true. Free to call.
`pixelift_batch`	credits	Submits up to 12 generating pixelift_ calls in one request (e.g. 6 video clips or 12 image variants) and returns the status and job id of each, in order. Asynchronous jobs come back as running: pass their ids to pixelift_wait (ids). Synchronous jobs run one after another inside the call; jobs that do not start within the time budget come back as not_started and can be resubmitted. Each job is charged like a direct call; the account limit of concurrent jobs still applies (too_many_active_jobs for the excess, nothing charged for those). Estimate the total with pixelift_estimate_cost first.
`pixelift_upload`	free	Stores a file the user attached in the chat (or a file from any https URL) in the Pixelift account and returns file_url to pass as image_url, video_url, media_url or audio_url to other pixelift_ tools. Pass exactly one of: data (the file bytes as base64, up to 10 MB) or url (https address, up to 35 MB). Formats: JPG, PNG, WEBP, MP4, MOV, WEBM, MP3, M4A, WAV (detected from the content). The file is private (only this API key's account can read it) and is deleted after 24 hours. Bigger files or a local file on disk: pixelift_create_upload. Free to call.
`pixelift_create_upload`	free	Returns a one-time upload_url valid for 15 minutes for a file up to 35 MB, for agents that can send HTTP requests (for example Claude Code: curl -T photo.jpg "<upload_url>"). Send the raw file bytes with HTTP PUT; after it succeeds, pass file_url to other pixelift_ tools. Same formats, privacy and 24 h lifetime as pixelift_upload. Free to call.
`pixelift_list_image_models`	free	Lists image models available for pixelift_generate_image with credits per image and strengths. Free to call.
`pixelift_list_video_models`	free	Lists video models available for pixelift_generate_video with credits per clip length, aspect ratios, sound and image-to-video support. Free to call.
`pixelift_get_balance`	free	Returns the credit balance of the Pixelift account behind the API key. Free to call.
`pixelift_estimate_cost`	free	Estimates credits for a pixelift_ tool call before running it. Pass the tool name and the same arguments you plan to use. Free to call.
`pixelift_get_generation`	free	Returns the status, progress and result URLs of a Pixelift job created by another pixelift_ tool. Free to call.
`pixelift_wait`	free	Waits up to 25 seconds for a job (id) or several jobs (ids, up to 12) to finish and returns the status and result URLs of each. With ids all jobs are checked in one request per poll, so waiting for 6 jobs costs the same as waiting for 1. Call again with the ids that are still running. Free to call.
`pixelift_cancel`	free	Cancels a queued or running asynchronous job of this account (video generation, talking presenter, lip sync, reframe, captions, video colorization, video upscale): the reserved credits are returned and the provider job is cancelled where possible. Calling it twice is safe (the second call reports the job as already cancelled, nothing is refunded twice). If the result arrived first, the job stays succeeded and credits stay charged (already_completed). Synchronous tools (images, voice-over, transcription) and product photoshoots cannot be cancelled: they finish within minutes. Free to call.
`pixelift_list_my_results`	free	Lists the account's recent jobs from the Pixelift library (newest first, 20 per page, up to 50) with status, tool, credits and result URLs, so an agent can reuse earlier outputs instead of regenerating them. Filter by status (e.g. succeeded), tool or source (mcp = created through this connector). Pass next_cursor from the previous call to get the next page. Free to call.

Presets (6)

Presets are MCP prompts: ready workflows that the agent carries out step by step with the tools above. In Claude Code type / and pick the preset, or run its command.

Meta ads product pack (1:1, 4:5, 9:16)

From one product photo: lifestyle and studio scenes, then every chosen scene in the three Meta ad formats (feed 1:1 and 4:5, Stories/Reels 9:16), upscaled for upload.

Command

/mcp__pixelift__meta_product_pack

Arguments

product_image_url: Public https URL of a clean product photo (packshot).
product_description: What the product is and who it is for, e.g. "vitamin C serum for women 30+".
scenes (optional): Optional: scenes to use (studio, marble, wood, kitchen, outdoor, dark, festive, beach) or a custom idea. Default: pick 3 that fit the product.

Static ad variants for A/B tests

From a short brief: 3 marketing angles × 3 headline variants as ready static ads with the headline rendered in the image, in 4:5 for the Meta feed.

Command

/mcp__pixelift__static_ad_variants

Arguments

brief: Product, audience, offer and tone, e.g. "eco water bottle, runners 25-40, -20% this week, energetic".
product_image_url (optional): Optional: public https URL of the product photo to keep in every ad.
language (optional): Language of the text in the ads, e.g. Polish. Default: the user's language.

UGC ad with a synthetic presenter (15-30 s)

From a product brief: a short script with a hook, a synthetic presenter who does not exist, a talking-head video with the script, burned-in captions, and the Reels 9:16, feed 1:1 and 4:5 formats.

Command

/mcp__pixelift__ugc_presenter_ad

Arguments

brief: Product, audience, key benefit, offer and tone, e.g. "collagen drink, women 35-50, visible skin effect in 4 weeks, -15% code SKIN15, friendly and honest".
product_image_url (optional): Optional: public https URL of the product photo (packshot) to show in the presenter's hand or next to them.
language (optional): Language of the spoken script and captions, e.g. Polish. Default: the user's language.
duration (optional): Target length in seconds, 15-30. Default 15.
presenter (optional): Optional: who the presenter should be (age range, gender, style), e.g. "woman around 40, natural look, kitchen at home". Default: matches the audience.

Localize a video ad into another language

From a finished presenter video: transcript, translation, a new voice-over in the target language, lip sync so the mouth matches the new speech, captions, and the Reels/feed formats.

Command

/mcp__pixelift__localize_ad

Arguments

video_url: Public https URL of the source ad (MP4/MOV/WEBM, presenter's face visible, up to 30 s).
target_language: Language to localize into, e.g. German.
source_language (optional): Optional: the language spoken in the source video. Default: detect.
notes (optional): Optional: brand terms to keep, tone, currency or offer changes for the new market.

Ad Resizer: one static ad in every format

From one finished static ad or product image: the same ad in 1:1, 4:5, 9:16, 16:9 or 1.91:1 without cropping. AI extends the background; the original text, logo and product stay pixel-exact.

Command

/mcp__pixelift__ad_resizer

Arguments

ad_image_url: Public https URL of the finished ad or product image (JPG, PNG or WEBP).
formats (optional): Optional: formats or placements, e.g. "4:5, 9:16" or "Meta feed, Stories, Google Display". Default: 1:1, 4:5 and 9:16.
notes (optional): Optional: what the new background must respect, e.g. "keep the extended area plain, no props".

Hook Multiplier: 3 new openings for one video ad

From one video ad: 3 versions with new hooks in the first 2-3 seconds and the rest of the ad unchanged, ready for an A/B test. Hook clips are generated with AI, then trimmed and joined with the original on the Pixelift server.

Command

/mcp__pixelift__hook_variants

Arguments

video_url: Public https URL of the source ad (MP4, MOV or WEBM, up to 35 MB and 2 minutes).
product (optional): Optional: what is advertised and to whom, e.g. "protein bar for runners 25-40".
hook_seconds (optional): Optional: how many seconds of the original opening to replace, 1-5. Default 3.
language (optional): Optional: language of any hook text and captions. Default: the user's language.
captions (optional): Optional: "yes" to burn captions of the speech into the final versions (charged per minute). Default: no.

Credits and files

Generating tools use credits from your balance. Balance, cost estimate and status tools are free.
Input files are passed as public https links. A result link can be the input of the next tool.
Results are kept for 30 days, so download what you want to keep.

Presets and examples

Ready prompts, presets and examples for agents are collected on a separate page.

Pixelift for AI agents